Two well-defined tools with comprehensive parameter schemas and detailed descriptions. Tool naming follows verb_noun convention. Schemas use Zod with proper type constraints and descriptions. However, output schemas are not explicitly documented, error handling is minimal, and composition is limited to two tools. The server demonstrates above-average schema rigor but lacks recovery guidance and formal output documentation.

• Document the output schema for each tool. For get_pedigree_documentation, specify: { "type": "object", "properties": { "content": [...], "metadata": { ... } } }. For generate_pedigree, document: { "type": "object", "properties": { "image": { "type": "string", "description": "Base64-encoded PNG or SVG XML" }, "width": number, "height": number, "format": enum } }.
• Add error handling with recovery guidance. Example: 'If generation fails with "Invalid dataset", check that all individuals have unique names (max 7 chars) and that all mother/father references resolve to existing individuals. Call get_pedigree_documentation() to review the structure.'
• Enforce parameter constraints in the schema using Zod conditionals. E.g., `z.object({ top_level: z.boolean(), mother: z.string().optional(), father: z.string().optional() }).refine(data => !(data.top_level && (data.mother || data.father)), { message: 'top_level=true requires NO mother/father refs' })`.
• Add a validate_pedigree_dataset tool that returns validation errors (missing parents, cycles, invalid references) before attempting generation. This enables the agent to fix data interactively.
• Consider adding a list_pedigree_conditions tool (read-only) that documents supported conditions/diseases and gene tests, enabling auto-complete and validation.
• Add a cancel/timeout mechanism for long-running renders. Document max execution time and what happens if generation exceeds it (e.g., returns partial result or timeout error).